Responsible by design
Gemma is designed with our Principles of AI at the forefront. In an effort to make Gemma pre-trained models safe and reliable, we used automated techniques to filter certain personal information and other sensitive data from the training sets. Additionally, we used deep and reinforcement learning from human feedback (RLHF) to align our instruction-friendly models with responsible behaviors. To understand and reduce the risk profile of Gemma models, we conducted robust evaluations, including manual red teaming, automated adversarial testing, and assessments of the model's capabilities for hazardous activities. These assessments are described in our Model card.
We are also publishing a new Responsible Generative AI Toolkit in collaboration with Gemma to help developers and researchers prioritize the creation of safe and responsible AI applications. The toolkit includes:
- Safety classification: We provide a new methodology to build robust security classifiers with a minimum of examples.
- Debugging: A model debugging tool helps you investigate Gemma's behavior and resolve potential issues.
- Advice: You can access best practices for model builders based on Google's experience developing and deploying large language models.
Optimized across frameworks, tools and hardware
You can fine-tune Gemma models on your own data to tailor them to specific application needs, such as synthesis or retrieval augmented generation (RAG). Gemma supports a wide variety of tools and systems:
- Multi-framework tools: Bring your favorite framework, with reference implementations for inference and fine-tuning on multi-framework Keras 3.0, native PyTorch, JAX, and Hugging Face Transformers.
- Multi-device compatibility: Gemma models work on the most common device types, including laptop, desktop, IoT, mobile and cloud, enabling widely accessible AI capabilities.
- Cutting-edge hardware platforms: We have partnered with NVIDIA to optimize Gemma for NVIDIA GPUsfrom data center to cloud to local RTX AI PCs, ensuring peak performance and integration with cutting-edge technology.
- Optimized for Google Cloud: Vertex AI provides a broad set of MLOps tools with a range of tuning options and one-click deployment using built-in inference optimizations. Advanced customization is available with fully managed Vertex AI tools or with self-managed GKE, including deployment to cost-effective infrastructure on GPU, TPU, and CPU from either platform.
Free credits for research and development
Gemma is designed for the open community of developers and researchers powering AI innovation. You can start working with Gemma today with free access to Kaggle, a free tier for Colab Notebooks, and $300 in credits for new Google Cloud users. Researchers can also apply Google Cloud credits up to $500,000 collectively to accelerate their projects.
To start
You can find out more about Gemma and access the quick start guides at ai.google.dev/gemma.
As we continue to expand the Gemma model family, we look forward to introducing new variations for various applications. Stay tuned for events and opportunities in the coming weeks to connect, learn and build with Gemma.
We're excited to see what you create!